How much water does ChatGPT use per prompt?

A single ChatGPT prompt uses roughly 0.28 mL of water on average, based on Google's 2025 published figure of 0.30 Wh per median AI query and the U.S. grid water intensity of 0.94 mL per Wh. Longer prompts or larger reasoning models can use several mL per prompt.

Does running AI prompts really affect the environment?

Individual prompts use small amounts — roughly 0.3 Wh of energy and a fraction of a milliliter of water. The environmental concern is scale: hundreds of billions of prompts per year means AI inference water use is climbing into the billions of liters annually, concentrated in water-stressed data-center regions.

How does ByteThirst measure AI water and energy use?

ByteThirst estimates prompt cost locally in your browser using token counts, a model-tier multiplier, Google's published 0.30 Wh baseline, and U.S. grid water and carbon intensity from EPA eGRID. The methodology is fully documented and uses only public, peer-reviewed sources.

Yes. The ByteThirst browser extension is free and works across 14 AI platforms including ChatGPT, Claude, Gemini, Perplexity, and Microsoft Copilot. A CLI tool is also free for developers who want per-API-call tracking. Premium ($3.33/mo) adds 90-day trends and spike alerts.

Which AI platforms does ByteThirst track?

ByteThirst tracks 14 platforms, including ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, Meta AI, Mistral Le Chat, DeepSeek, Grok, Google AI Studio, Lovable.dev, and Bolt.new. New platforms are added based on user requests and market share.

Does ByteThirst send my prompts anywhere?

No. All measurement happens locally in your browser. Prompt contents are never transmitted to ByteThirst servers or any third party. The extension only counts tokens and applies the public methodology — your conversations stay private.

ByteThirst: AI Water & Energy Tracker — See the Cost of Every AI Prompt

Compare AI models. See how to reduce the footprint.

Pick a sample prompt, pick a model, see the estimated water, energy, and CO₂. Click the ⓘ icons to read how each is calculated.

input tokens · output tokens

Your model Compare against

Your model

Energy: ~ Wh
Water: ~ mL
CO₂: ~ g

Comparison

Energy: ~ Wh
Water: ~ mL
CO₂: ~ g

~% lower energy than primary (within-product, holding methodology constant)

Methodology v2.1 · Read the full methodology

See savings like this on every prompt you send.

ByteThirst tracks your AI usage across 14 platforms — ChatGPT, Claude, Gemini, and 11 more. Free includes daily tips on lower-impact alternatives from our standard library. Premium unlocks an expanded tip library, session-level recommendations, cross-platform trends, and team rollups.

Add to Chrome Try Premium free for 7 days

Card on file via Stripe — cancel any time before day 7.

You’re Probably Underestimating Your AI Footprint by est. 130×

A single vibe coding session generates dramatically more estimated environmental impact than a typical text chat.

Text Chat

est. 0.30 Wh

≈ 0.28 mL water per query

ChatGPT, Claude, Gemini

Vibe Coding

est. 41 Wh

≈ 38.5 mL water per session

Lovable.dev, Bolt.new

Source: Couch, S.P. (2026), “Electricity use of AI coding agents”

How It Works

Four steps. No setup. No learning curve.

Step 1

Install & Forget

Add ByteThirst to Chrome. It runs silently in the background — no setup, no account, no interruptions.

Step 2

See Your QueryWeight

Open the popup anytime to see your daily and weekly estimated AI footprint — water, energy, and CO₂ with standard volume conversions.

Step 3

Understand & Share

Generate weekly impact cards and share your estimated usage on social media. Earn achievement badges along the way.

Step 4

Go Team

Upgrade to Teams and see your organization’s collective QueryWeight. Set shared goals and build awareness together.

Estimates Across 14 AI Platforms

From everyday chat to vibe coding — ByteThirst estimates the environmental cost across the AI tools you use most.

Chat

ChatGPT Claude Gemini Copilot Perplexity Mistral You.com

Code

Lovable.dev Bolt.new Google AI Studio

Specialized

Poe HuggingChat NotebookLM Figma AI

What You Get

Everything you need to understand your AI’s estimated environmental footprint.

Monthly Annual

Free Tier

Free — forever

Everything you need to understand your AI footprint.

Daily & weekly QueryWeight dashboard
Water, energy & CO₂ estimates with ranges
14 AI platforms including code builders
Volume conversions (tsp, tbsp, fl oz, cups)
6 achievement badges
1 shareable social impact card per week
Community Benchmarks (opt-in on all tiers)
CLI tool for terminal-based AI tracking
Local-first — no account required

Chrome Browser Extension

Download Now

Or install the CLI

npm install -g bytethirst

Premium

$3.33 /month

$39.99 billed annually

Advanced QueryWeight analytics for impact-conscious users.

Everything in Free, plus:
Historical trend analytics (30/60/90-day charts)
Unlimited social impact cards
Usage spike alerts
Code generation impact breakdowns
15+ achievement badges with streak tracking
Early access to new platform support
Unified Chrome Extension and CLI Dashboard

Start Free Trial

7-day trial. Card on file via Stripe — cancel any time before day 7.

Collaborative

$13.33 /seat/month

$160 per seat billed annually

Collaborative QueryWeight insights for your team. 5-seat minimum.

Everything in Premium, plus:
Team QueryWeight dashboard
Team goals and challenges
Admin controls & insights
ESG-ready quarterly reports

Start Team Trial

7-day trial. Card on file via Stripe — cancel any time before day 7.

Enterprise (100+ seats) — SSO, API access, and Scope 3 reporting.

Contact us for custom pricing →

View full pricing comparison →

What Is a QueryWeight?

QueryWeight is the estimated environmental cost of a single AI interaction — expressed in water, energy, and carbon.

💧

Water

Data centers use water for cooling. Every AI query contributes to an estimated water footprint — from fractions of a milliliter to tens of milliliters for code generation.

⚡

Energy

AI models consume energy with every inference. A standard text query uses an est. 0.30 Wh — while a code generation session can use est. 41 Wh.

🌎

Carbon

Energy generation produces CO₂. ByteThirst converts estimated energy use into carbon emissions using the US grid average of est. 0.39 kg CO₂/kWh.

Transparent by Design — The QueryWeight Pipeline

Every number has a source. Every estimate shows uncertainty.

Your AI Query Token Estimation Energy (Wh) Water (mL) CO₂ (g)

est. 0.30 Wh

Energy per standard query

Source: Epoch AI, 2025

est. 0.94 mL/Wh

Water-to-energy ratio

Source: OpenAI, 2025

est. 0.39 kg/kWh

US grid carbon intensity

Source: EPA eGRID, 2023

est. 130×

Code generation vs. text chat impact

Source: Couch, 2026

Read the full methodology →

Privacy Is the Foundation

We built ByteThirst to be private by default, not private by promise.

What we NEVER collect

Your conversations (prompts or responses)
What topics you discuss
Exact timestamps (only the date)
URLs you visit
Your identity (free tier)
Browser fingerprint

What stays on your device

Number of queries per day (by platform)
Estimated token counts
Model type (e.g., GPT-4o, Claude Sonnet)
Date (no time)

Local by default. All QueryWeight calculations happen on your device. Premium transmits only your email and subscription status. Teams shares daily aggregate QueryWeight per platform with your admin. On any tier, you can opt in to anonymous Community Benchmarks — no personal info is ever collected.

Read the full privacy policy →

Also Available as a CLI

Estimate the environmental impact of terminal-based AI tools — Claude Code, Aider, GitHub Copilot CLI, Cursor, and OpenAI Codex.

Terminal

——— ByteThirst™ Session Summary ———————————————————————————
  💧 est. 0.1 mL
  ⚡ ≈ 0.06 Wh energy
  🌍 ≈ 0.02 g CO₂
  Queries: 1 · Platform: claude · Model: sonnet (standard)
  QueryWeight™ values are estimates.
————————————————————————————————————————

Learn More about the CLI